Human Perception-Oriented Enhancement and Smoothing for Perceptual Video Coding

We propose a human perception-oriented processing method that is readily applicable to human perceptual video coding. While existing methods focus on removing perceptual redundancy, the proposed method aims to enhance image quality in the region of interest while suppressing non-perceivable signals to reduce perceptual redundancy at less interesting regions and smoothing unpleasant noise patterns in motion-blurred regions. To implement the proposed scheme, we propose a DCT domain image enhancement filter, a DCT domain just noticeable difference (JND)-based Gaussian filter, and a DCT domain motion blur smoothing filter. The enhancement filter is based on the proposed signal direction adaptive multiband energy scaling and enhances local image signals without evoking ringing and blocking artifacts. The JND-based Gaussian filter reduces perceptual redundancy using the estimated JND values that partly depend on the distance from an eye-focused point. Motion-blurred image signals are smoothed by the DCT domain Gaussian filter. To demonstrate the effectiveness of the proposed method, it is applied to video sequences before encoding them. The experimental results verify that the proposed method improves perceptual video quality consistently for various sequences and at diverse bandwidths.